Metagenomic Barcoding

March 1^st, ’23

Overview

1 | background

2 | questions

3 | approaches

4 | methods

5 | results

6 | future

7 | conclusions

Cercocarpus

Team

Jane Ogilvie (ecological fieldwork, design)

Emily J. Woodworth (pollen morphology, and microscopy)

Sophie Taddeo (geo-spatial, statistics)

Paul CaraDonna (little bees/big picture(s))

Jeremie Fant (all things molecular and tied together)

i.e. an in-house production.

the world is big - 1.1

Primary Roads

… really really big - 1.2

5 sampling seasons (May - October)
3 person crews
2 partial support personnel
281 plots
area of inference: ~ 900,000 acres
0.363% of Bureau of Land Management administered land

FQI calculated from AIM

funding opportunties - 1.3

## how do we sample the planet? - 1.4

plant species in ecology - 1.6

mis-identification is very common
mis-identification can lead to nebulous understandings
mis-identification can lead to mis-management

Cirsium scariosum

insects species in ecology - 1.7

Macro Invertebrates
- stream ecology bio indicators
  - mayflies, caddisflies, stoneflies
Coleoptera
- soil contamination by metal
from bio-indicators to foci?

Macroinvertebrates.org

from organisms to interactions - 1.8

Bullet 2
Bullet 3

Solidago spathulata & Megachile wheeleri, by A. Litz

metabarcoding - 1.10

Barcoding
- molecular identification of tissue from a single organism
Metabarcoding
- molecular identification of organisms present in a mixed substrate

Five Astragali

barcodes - 1.11

Kingdom: Animal,
- COI (Cytochrome c OxIdase)
- holding it’s own in Fungi
Kingdoms: Fungi + Plant
- ITS (Interal Transcribed Spacer)
- holding it’s own in Fungi
Kingdom: Plant
- ITS, rbcL, matK, trnH-psbA
- not holding much of anything

new barcodes for plants? - 1.12

genomics
- low cost
- high coverage
- PCR free?
reference library ?
- old barcode library in development for nearly 20 years
- Kew PAFTOL
angiosperms 353

2 - questions

scale - from plots to continents?

many questions will be approached using two perspectives
- bottom up i.e. plot based data collected by Jane
- top down i.e. computer based data generated by me
fine scale data serving to as ground truth to the computer generated models

can we predict what is flowering in time & space? - 2.1

which species are present in an area?
when are these species flowering in an area?
diverse clades provide challenges for identification
species often diverged in ecological traits

Rhododendron sp. Hengduan Mtns., by Qin Li

do a353 work as barcodes? - 2.2

‘universal’ markers for phylogenomics
usable in all flowering plant clades
first comprehensive genus level phylogeny of flowering plants
shoot the moon; meta genomics first

are a353 semi-quantitative? - 2.3

Do the number of sequence reads reflect the amount of biological material in a sample?

3 - approaches

predict what is flowering; Time & Space - 3.1

no longer any funding for floristics; few Floras maintained, fewer written
essentially no funding remains for alpha taxonomy
little to no funding natural history
how do we monitor ecological shifts under climate change?
- geographic ranges
- flowering time
back to the sheets!

FPNW 2^nd

custom sequence databases; a353 as barcodes? - 3.2

reduce number of species present in database
- reduce computational requirements
- increase likelihood of relevant matches across loci
- reduce false positives for semi-quantitative inference

queen bee pollen loads: a353 as barcodes? - 3.3

DNA extracted from corbiculae loads
- a ‘pollen basket’ for holding grains collected grains
variable in size, but generally many tens of thousands of grains

USFWS

identify pollen grains; a353 semi-quantitative? - 3.4

4 - methods

field work
spatial
temporal
morphologic
laboratory
bioinformatic
post-classification

study system & field work - 4.1

pollen morphological identification 4.2

Develop grain reference library
Score traits
write keys
Emily Woodworth talk at ‘Not Another Specimen’ ASPT @ Botany 2021

workflow

pollen reference library 4.2.1

ca. 110 species
60 novels species added
shared
many more species to add to key!! (60 +, mostly un- sampled families)

pollen corbiculae loads 4.2.2

aliquot from same sample used for molecular
stained by fuchsin jelly with stirring
transects
rarefaction curves
- richness
- abundance

Corbiculae Sample

molecular barcoding 4.3

Angiosperms 353…

spatial analysis 4.3.1

2-stage approach
- 1^st: distance search of records from museums & plot based data (e.g. Forest Servce)
- 2^nd: species distribution modelling

plant species, distribution modelling 4.3.1.1.

develop a candidate species list for barcoding
download all herbarium records from a distance exceeding the study area
compare to known species at field site
logistic regression
bootstrapped samples of records

species distribution modelling - 4.3.1.2

gather records from aggregator websites - BIEN
prune records and model
GLM & GAM, RF & boosting
ensemble individual models
talk for analytical toolkit (PBC 470)

a single species

sdm evaluations - 4.3.1.2

in pipeline, True skill statistics
- works well over wide range of occurrence records

temporal modelling - 4.3.2

reduce herbarium records to study domain
thin records to analogous ecoregions
trim start/end records
identify major phenological cues, subset records to similar areas

temporal modelling subset - 4.3.2.1

SPATIAL SUBSET PICTURE

temporal modelling distributions - 4.3.2.2

barcode references library - 4.4

genomics work - 4.4.1

Plant Reference Library
- herbarium & silica dried
- CTAB, some DNEasy
Pollen Extraction
- ‘novel’ CTAB / SDS extraction
Both
- clean up Cytiza, size selection SPRI
- enzymatic fragmentation

plant genomic reference dna - 4.4.2

38 species to sequencing
13 species duplicate
24 silica gel dried, 14 herbarium leaf tissue (RM, ID, IDS)

tissue from Rocky Mountain Herbarium

pollen genomics dna - 4.4.3

54 Initial samples for extraction
44 samples underwent all steps and were analyses

hyb-seq

barcoding informatics - 4.4.4

trimmomatic, remove tags, select sequences > 31 bp in length
Kraken - qualitative identification
Bracken - quantitative identification
BLAST followup

metabarcoding - 4.5

sequence database generation - 4.5.1

Kew Tree of Life ~ ### taxa
US ~ xx TAXA

database

sequence assignment - 4.5.2

semi-quantitative evidence 4.5.3

5

results

field work

723 floral visitation observations (!)
36 unique plant species involved
64 corbiculae loads from Queens

sdm candidate species ????

downloaded some 112k records
mostly trees from forestry surveys
bootstrap re-sampled to reduce effects of collection ‘hotspots’
non-present taxa begin nearly immediately…
real occurrences taper off quickly

database

sdm evaluations - computational - 5.2

Logistic regression assessing accuracy of SDMs; witheld data
Metric	Value	Metric	Value
Accuracy (Training)	83.75	F-Score	0.84
Accuracy (Test)	84.00	AUC	0.92
Recall	81.03	Concordance	0.92
True Neg. Rate	86.97	Discordance	0.08
Precision	88.04	Tied	0.00

sdm evaluations - 5.3.1

able to compare to a localized vascular plant checklist
- not everything w/ vouchers…
- not everything w/ vouchers…
able to remove nearly all species from the upper (alpine) & lower (sagesteppe) life zones

	ml	lm
ensembles	493	473
true +	362	286
true -	33	55
false +	64	41
false -	34	93

sdm evaluations - 5.3.2

We were interested in comparison to the Valleys.
Plot Level, 117 species total (109 eligible for modelling…)
- ML: 105 (89.7% (96.3%))
- LM: 102 (87.2% (93.5%))
Able to detect virtually all species recorded on plot

coarse phenological modelling - 5.4.1

strong agreement between first and peak flower periods with historic data
good agreement between last flower date
no agreement with duration! - species do not ‘line up’

flower dates

coarse phenological modelling - 5.4.2

similar results with weekly data across all field sites combined
tau values lower than over longer term data

flower dates

metabarcoding - 5.5

sequence database generation - 5.6

found existing data for 130 species on NCBI - SRA
novel sequence data for 25 species, varying number of loci
whole ‘ring’ to be completed within the year

Species in Sequence DB

sequence assignment - 5.7 - I

trimmomatic (discard short reads)
Kraken (many false positives)
Bracken (many many false positives)
Blast (fewest false positives)

Three Initial Networks

We utilized trimmomatic to trim unique identifiers which are attached to all samples for metagenomic sequencing, which have been shown to give false positives, and to remove short reads when we were at it. Essentially Kraken requires reads at least 30 base pairs long and these were removed at this stage. We initially planned on using a single algorithm for classifying our sequence reads. Kraken, kraken is a very fast sequence alignment algorithm which has enormous popularity in the microbiome world, and the application of metagenomics there.

Subsequent to Kraken, which will classify reads to the lowest unambiguous taxon which it can, i.e. it can return matches for ‘Rosaceae’. Bracken ‘Bayesian Reestimation of Abundance with KrakEN’, is an extension of kraken which seeks to classify all reads from higher to a terminal taxon. It also is supposedly quite good at estimating abundances of reads in samples. We found that while Bracken would place all results, we were very skeptical of them.

Accordingly, we decided to give BLAST, the golden standard of sequence alignments a go. We aligned all sequences which were used by Kraken, so essentially kraken became an intermediate filter, to BLAST. we were considerably happier with the results from Blast being closer to what field data told us.

sequence assignment - 5.7 - II

Post classification of Sequences via Taxonomy and Ecology, top 15 most abundant reads
Condition	No. Class.	Prcnt. Class.	Total Seqs	Rank
A	143	21.0	32.0	Species
B	205	30.1	10.5	Species
C	5	0.7	0.4	Genus
G	29	4.3	7.8	Species
H	280	41.2	47.9	Genus
None met	18	2.6	1.4	Multiple

sequence assingment - 5.7 - III

Naive BLAST from custom databases 26% accuracy
post-classified BLAST using temporal filters to create genera monogeneric in space and time 44%
BLAST, creates many false positives

sequence assingment - 5.7 - IV

conceptually similar to the automated process
utilized high resolution occurrence and phenology data
utilized morphological and molecular data
no linear operation or rule of precedence
classified all sequences to species

semi-quantitative evidence - 5.8

…
some relationship exists
requires further work by someone else

Counted grains

final floral feeding structure - 5.9

6

Discussion

conservation implications - 6.1

a hidden plain text paragraph in the discussion.
collaboration with Ken Holsinger, Jedd Sondergard @ BLM Montrose

conservation implications - 6.1 - II

historic vegetation treatment removals
- Delphinium, Astragalus & Oxytropis
altered fire cycle
- Delphinium, Mertensia
stream channelization / wetland removal
- Mertensia
seed species, where missing
allow return of historic fire cycle, when appropriate
reintroduce beavers, or if unavailable than analogs

7

future

metabarcoding; computational approaches - 7.2

qualitative:
- search for variable loci
- flanking regions and pop gen
quantitative:
- read re-assignment based on phylogentic distance
- read re-assignment in bayesian framework

metabarcoding; new data sets? - 7.1

(more) West Slope
artificial mixtures
- leaf tissue (counted cells)
  - easier to collect
- pollen loads (counted grains)
  - more popular
Gunnison Sage-Grouse scat
- BLM habitat assessment data
- opportunistic collections

bombus; trends in perennial bunchgrasses - 7.3

notable declines in bunch grasses over the last 40 years
especially pronounced in montane areas
model conversion of meadows/Aspen stands to conifer forest
use existing data to model reductions in grass cover
view historic grazing renewals to monitor (loosely) reduction in grass diversity
forked from Ken Holsinger

8

conclusions

promising

acknowledgements

two super fantastic technicians for the field seasons I worked during school, made my life SO easy
find them funding and recruit them, and then give me a finders fee.

Dani Yashinowitz B.S. (Yellowstone National Park, botanist & crew lead, Whitebark Pine Surveys (!!!))
Hannah Lovell B.S. (Telluride Mountain Resort, and in search of work)

acknowledgments

Employment: Yingying Xie, Josh Scholl, Sam Isham, Kelly McMillen, Kay Hajek, Linda Vance, Cassandra Owen, Ken Holsinger

Project: Nyree Zerega, Pat Herendeen, Hilary Noble, Zoe Diaz-Martinez, Angela McDonnell, Elena Loke, Ian Breckheimer, Ben Legler, Ernie Nelson, Charles (Rick) Williams, D. Knoke, L. Brummer, J. Boyd, C. Davidson, I. Gilman, M. Kirkpatrick, S. McCauley, J. Smith, K. Taylor, & C. Williams. David Giblin, Mare Nazaire, Sarah Burnett, Lauren Price, T.C.H. Cole, Eliot Gardner.